---
title: Car Analysis Advisor (Multi‑Image, Deterministic)
emoji: 🛠️
colorFrom: purple
colorTo: indigo
sdk: gradio
sdk_version: "4.36.1"
app_file: app.py
pinned: false
license: mit
tags:
  - automotive
  - computer-vision
  - service-advisor
  - gradio
---

# Car Analysis Advisor (Hugging Face Space)

An end‑to‑end **car analysis AI** that ingests **multiple images** of a vehicle and outputs **precise, actionable recommendations**:
- **Model** (zero‑shot CLIP baseline; optional fine‑tune)
- **Color** (dominant body color with named snapping)
- **Issues** (mechanical/aesthetic) → **deterministic** final decisions
- **Exact price estimate** (parts + labor + region multipliers)
- **PDF & JSON** export

No probability scores are exposed—only clear decisions & totals.

## Run locally
```bash
pip install -r requirements.txt
python app.py
```

## Train (optional, improves accuracy)
See `training/train_classifier.py` for a ViT fine‑tuning script.
Dataset CSV:
```csv
image_path,label
images/nexon_001.jpg,Tata Nexon
```
Then:
```bash
python training/train_classifier.py --data_root data/your_dataset --annotations data/your_dataset/annotations.csv --out_dir checkpoints/vision
```

## Pricing configuration
- `configs/parts_catalog.yaml` — parts & standard labor hours per issue
- `configs/regions.yaml` — labor rates & part multipliers per region (example values provided)

## Deterministic decisions
Internally, scores are computed but **final outputs are rule‑based** (top evidence & thresholds) to provide **unambiguous** issue lists and **exact** costs.

## Exports
- `exports/report.pdf`
- `exports/report.json`